Scholarlink

<p>A renowned topic in computer vision and Natural Language Processing is Image Captioning (IC). Numerous works focused on automated systems. However, these works often failed to cover all the information in images and the sentiment of captions. Therefore, an automated IC system is presented using Positional and Dual Attention with Alignment-based Deep Convolutional Long Short Term Memory (PD2A-DCLSTM). Initially, the input image undergoes preprocessing by utilizing the Lorenz Curved Contrast Limited Adaptive Histogram Equalization (LCCLAHE) for contrast enhancement. Objects in the image are then detected using Bayesian You Only Look Once (B-YOLO). Kendal Rank Correlation (KRC) is computed between segmented objects, and similar objects are grouped using Secant Root Functional Affinity Propagation (SRFAP). Key points are extracted from segmented objects using Scharr Scale Invariant Feature Transform (SSIFT). Features are then extracted from pixels of pre-processed images, key points, and grouped objects. Important features are selected using the Differential Function-based Zebra Optimization Algorithm (DF-ZOA). These selected features and pre-processed images are input to the PD2A-DCLSTM. The sentiment score of the generated caption is evaluated and contrasted with the labelled caption. The similarity score is computed, and the sentiment feedback is sent to the PD2A-DCLSTM. In experimental analysis, the proposed model attained a 0.887 BiLingual Evaluation Understudy (BLEU) score.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced image captioning with positional and dual attention using deep convolutional long short-term memory and emotional feedback mechanism

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • A. Ranjith Gnana Suthakar,
  • B. J. Sandesh

摘要

A renowned topic in computer vision and Natural Language Processing is Image Captioning (IC). Numerous works focused on automated systems. However, these works often failed to cover all the information in images and the sentiment of captions. Therefore, an automated IC system is presented using Positional and Dual Attention with Alignment-based Deep Convolutional Long Short Term Memory (PD2A-DCLSTM). Initially, the input image undergoes preprocessing by utilizing the Lorenz Curved Contrast Limited Adaptive Histogram Equalization (LCCLAHE) for contrast enhancement. Objects in the image are then detected using Bayesian You Only Look Once (B-YOLO). Kendal Rank Correlation (KRC) is computed between segmented objects, and similar objects are grouped using Secant Root Functional Affinity Propagation (SRFAP). Key points are extracted from segmented objects using Scharr Scale Invariant Feature Transform (SSIFT). Features are then extracted from pixels of pre-processed images, key points, and grouped objects. Important features are selected using the Differential Function-based Zebra Optimization Algorithm (DF-ZOA). These selected features and pre-processed images are input to the PD2A-DCLSTM. The sentiment score of the generated caption is evaluated and contrasted with the labelled caption. The similarity score is computed, and the sentiment feedback is sent to the PD2A-DCLSTM. In experimental analysis, the proposed model attained a 0.887 BiLingual Evaluation Understudy (BLEU) score.